A new technique developed by scientists in Spain enables the systematic search for elusive allosteric binding sites on proteins.

Applying the method to map allosteric sites of two pharmacologically important proteins, the researchers show both that these sites are more common than previously thought, and that rather than turning off or on when engaged, they can be modulated.

"The approach could be a game changer for drug discovery, leading to safer, smarter and more effective medicines," said Andre Faure, first author of the paper describing the method published in Nature, on April 6. "It enables research labs around the world to exploit vulnerabilities in any protein, including those previously thought to be undruggable," he said.

Allosteric effects occur when a molecule binding to the surface of a protein initiates changes at a distant site, regulating its function by remote control. Although some allosteric sites discovered serendipitously have proven to be important targets, there previously was no reliable way to find them.

That is because proteins change conformation in the presence of an incoming molecule, exposing hidden pockets within the surface that are potentially allosteric, but which cannot be identified with conventional structure determination techniques.

Allosteric regulation is central to signal transduction, transcriptional regulation and metabolic control. In addition, many disease-causing mutations, including those that drive tumor growth, are pathological because of their allosteric effects.

At the same time, the specificity of allosteric sites makes them better drug targets than the orthosteric sites that are conserved across protein families.

The understanding that allosteric communication between distant sites is so central to biological regulation -- and the frustrating lack of tools to characterize this interchange -- led to it being labeled 'the second secret of life.'

The researchers postulated that global maps of allosteric communication could be generated for protein binding domains if the effects of all mutations on binding affinity could be quantified, since any mutation altering binding affinity, but not directly contacting a ligand, must be having an allosteric effect.

The problem they had to crack is that changes in affinity cannot be inferred simply by quantifying changes in binding to an interaction partner.

That is because changes in the molecular phenotype on binding to an interaction partner can be caused by many different changes in the underlying biophysical properties.

"To quantify the effects of mutations on binding affinity and to globally map allosteric communications, these ambiguities must be resolved," the researchers say.

They devised a strategy to do this, which they call multidimensional mutagenesis. It takes advantage of massively parallel deep mutational scanning to quantify the phenotypic effects of thousands of perturbations, which allows the effects of mutations to be quantified for multiple molecular phenotypes in multiple genetic backgrounds.

The method resolves ambiguities where a number of biophysical changes could account for an observed mutational effect and allows the inference of the in vivo biophysical effects of mutations.

The researchers then used neural networks to fit thermodynamic models to these experimental measurements, thus inferring the underlying causal changes in free energy.

By applying that to two of the most common protein interaction domains – the C-terminal SH3 domain of the human growth factor receptor-bound protein 2 (GRB2), which binds to GRB2-associated binding protein; and the third PDZ domain from the adaptor postsynaptic density protein 95 (PSD95), which binds to the C-terminus of CRIPT – the method provided "near complete" views of their free-energy landscapes and made it possible to build global maps of allosteric mutations.

Co-author Ben Lehner, coordinator of the systems biology program at the Center for Genomic Regulation of the Barcelona Institute of Science and Technology, describes this as a "brute force experiment", in which generating thousands of versions of a protein, each just one or two amino acids different from another, makes it possible to build an overall picture of how it functions.

"It's like suspecting a faulty spark plug, but instead of only checking that, the mechanic dismantles the whole car and checks it piece by piece," he said.

"This method will have broad applications in biotechnology," say Debora Marks of the Department of Systems Biology at Harvard Medical School and Stephen Michnick of the Department of Biochemistry at the University of Montreal, in a commentary accompanying the paper.

The resources used are similar to those available at most university or industry laboratories and the method is easy, "making it as accessible to geneticists as it is to biophysicists," Marks and Michnick say.

Faure et al. say systematic maps of spatial information transfer in proteins, and experiments to identify modifiers of this transport, will open the door to greater understanding of allosteric regulation. Further scaling up could make it possible to predict the properties of proteins from their amino acid sequences, they add.

It took 50 years for computational methods to have the power needed to resolve a protein's structure from its amino acid sequence. Lehner believes the same could be achieved in terms of predicting the properties of proteins in 10 years.

"Now we know this general approach is likely to work we can apply it to many other fundamental problems in biology – such as predicting protein stability, binding affinities and allostery. But for these other problems the required quantitative data do not exist yet – we need to produce them," he said.

"The most important challenge is how to efficiently generate these experimental datasets so that solving these problems does not take another 50 years. We think that, with a sensible investment in data production, it could be possible to 'solve' allostery and many of these problems within 10 years," Lehner told BioWorld Science.

Generating allosteric target maps for therapeutic proteins should help accelerate allosteric drug discovery. "Not only are these potential therapeutics sites abundant, there is evidence they can be manipulated in many different ways. Rather than switching them on and off, we could modulate their activity like a thermostat," Faure said.

The method itself already provides good evidence that these novel allosteric sites can be modulated. "The thousands of amino acid substitutions that we introduce in the two assays [...] change the local chemistry at those positions in the protein and we measure their effects," Faure told BioWorld Science.

The scientists at the Center for Genomic Regulation are now applying their method to cancer proteins and cell surface receptors targeted by many approved drugs. "We expect that these projects will provide important mechanistic insight into how these proteins work and also identify new sites that could be targeted to develop new therapies," said Lehner. As one example, a colleague has built a map of an important cancer gene, which he said, "looks really interesting."